What we automated stopped where a patient’s appeal began.

A prior authorization denial arrives as a three-page PDF with one sentence telling a patient their treatment was refused. The 60-day appeal window starts that day. Most patients never find the sentence.

Brivon offers help with those appeals while keeping advocates, who handle one case at a time, from burning out.

Worked with: an AI product manager on requirements and the build; 6 people interviewed.

I audited the market and interviewed 6 people, including patients managing chronic pain. Two advocates stopped mid-sentence in testing when the AI got a case wrong.

Insurers now use AI to deny claims faster than a person could review them. Patients have nothing comparable.

I rejected care navigation, billing disputes, and a general health platform. Care navigation had the strongest emotional pull, but its win condition was too diffuse to judge in weeks. A binary outcome also made a success fee possible without a track record.

The cost is a narrow entry point that would make a small business if adjacent workflows never opened.

Phase 0 covers AI-assisted intake and classification, manual advocate work, and a case timeline. I rejected a full copilot that drafts appeals, even with advocate review, though it was reachable by week two.

I accepted the cost: advocates write appeals by hand on a success fee that only pays when cases close. A misclassified denial routes to the wrong template, and a missed deadline can postpone surgery. An advocate who catches the AI wrong twice stops trusting it. Nothing leaves the building without an advocate reading it first. That gate is the product.

The queue puts a deadline expiring today above a review that can wait until Thursday. The depth-first alternative meant eight context switches across eight cases before the day began. Testing showed advocates also wanted their workload on screen, so I added a session summary.

Profiles lead with patterns such as “15 of 17 denials at this insurer for this condition reversed on first appeal,” not credentials. Patient testimonials would expose people. A tester said the acceptance rate was what separated the product from a directory: “So they’re not just listing everyone.”

I cut AI-drafted appeals, proactive deadline monitoring, employer dashboards, and any browsing surface. Each added risk or needed data the pilot would not have.

The brand took three directions. Clinical Authority read like a hospital’s internal tool. Warm Advocacy was called a meditation app. Earned Confidence came from asking what someone looks like when they have done this before and won.

Phase 0 passes if 70% or more of cases reach a clear outcome signal inside the case lifecycle. I would also watch advocate throughput on a half-automated system. None of this has happened. These are pilot hypotheses, not results.